SCIENCE TRANSLATIONAL MEDICINE | FOCUS
ALZHEIMER’S DISEASE
Endosomal recycling reconciles the
Alzheimer’s disease paradox
Scott A. Small1* and Gregory A. Petsko2
A hub-and-spoke model with endosomal recycling as the hub can reconcile the pathogenic contribution of
amyloid precursor protein to Alzheimer’s disease.
THE ALZHEIMER’S DISEASE PARADOX
The hallmark histopathologies of Alzheimer’s
disease (AD) have remained essentially the
same as those described by Alois Alzheimer
over a century ago, except that we now have
a deeper understanding into their molecular
and cellular underpinnings. The extracellu-
lar plaques of amyloid pathology chiefly
comprise the A peptide, a proteolytic frag-
ment of the amyloid precursor protein (APP)
that is primarily produced and secreted by
neuronal endosomes. The pathogenic neuro-
fibrillary tangles found inside neurons in
the AD brain comprise the microtubule-­
binding protein tau. Tau pathology begins
in neurons, but it is now known that tau can
be actively secreted, representing one of the
earliest pathological events. AD is a neuro-
degenerative disease, but before the neuro-
degenerative process causes neuronal cell
death, it is preceded by synaptic pathology
that is characterized by loss of dendritic
spines. In contrast to these neuronal histo-
pathologies, there is an additional patholo-
gy that is localized to cells called glia in the
brain. Recent studies have shown that AD’s
hallmark glial pathology appears to primarily
affect a subset of glia, the microglia. A char-
acteristic of microglial pathology in AD is
their diminished ability to phagocytose ex-
tracellular proteins and perhaps debris.
A hypothesis governing the relationship
of these four AD pathologies—amyloid, tau,
synaptic and microglial—was proposed when
genetic mutations linked to rare autosomal-­
dominant, early-onset AD were identified
and when subsequent investigations showed
that these mutations all belonged to the same
biological pathway. The hypothesis, now
called the amyloid cascade hypothesis, was
first formulated when mutations in APP
were found in individuals with the early-­
1
Alzheimer’s Disease Research Center, Department of Neurology, Columbia University, New York, NY 10032,
USA. 2Ann Romney Center for Neurologic Diseases, Department of Neurology, Brigham and Women’s Hospital
and Harvard Medical School, Boston, MA 02115, USA.
*Corresponding author. Email: sas68@cumc.columbia.edu
Small and Petsko, Sci. Transl. Med. 12, eabb1717 (2020)
onset inherited form of AD. It then gained
momentum when additional mutations were
isolated in genes that were ultimately called
presenilin1 (PSEN1) and presenilin2 (PSEN2),
because they were shown to encode proteins
involved in the enzymatic cleavage of APP.
Biochemically, all three mutations directly
affect the proteolytic processing of APP,
and all three were found to worsen amyloid
pathology. Like any good scientific hypothesis,
the amyloid cascade hypothesis is anchored
by two specific and testable assumptions.
First, extracellular amyloid pathology in
the brain—the plaques observed by Alois
Alzheimer or the soluble cloud of amyloid
fragments that surrounds them—was stipu-
lated to represent the upstream cause of the
disease. Second, in a linear series of causal
steps, extracellular amyloid pathology was
proposed to secondarily trigger AD’s other
three pathologies. In this way, extracellular
amyloid was postulated to be the upstream
cause of AD, setting into motion a domino
effect of the other three pathologies. Although
the mutations in APP or its cleaving enzymes
are not found in the common late-onset
form of AD, both forms of the disease share
the same four defining core pathological
features, and so, the hypothesis has been
proposed to be generalizable to both.
The amyloid cascade hypothesis mobilized
the pharmaceutical industry to develop a range
of drugs and to complete over 100 clinical
trials, all targeting amyloid pathology. Un-
fortunately for patients with AD, nearly all
have failed. The minimal clinical efficacy of
these candidate drugs has received the most
attention in calling the amyloid cascade
hypothesis into question. In truth, the com-
plexity of clinical trials that rely on cognitive
measures for a slowly progressing disease
must be recognized, and there can always be
2 December 2020
Copyright © 2020
The Authors, some
rights reserved;
exclusive licensee
American Association
for the Advancement
of Science. No claim
to original U.S.
Government Works
post-hoc practical reasons for failed clinical
end points. Given that many of the most
Downloaded from http://stm.sciencemag.org/ at CARLETON UNIVERSITY on December 2, 2020
recent clinical trials tracked AD pathologies
in living patients and in postmortem brains,
these trials can also function as experimen-
tal medicine studies from which scientific
conclusions can be inferred. More than
showing little meaningful clinical benefit,
perhaps the most telling finding from these
studies is that despite clearing amyloid patholo-
gy, even in relatively early stages of the disease,
the other pathologies have often been found
to continue unabated [e.g., (1)]. Extracellular
accumulation of amyloid pathology can at some
level be cognitively detrimental, and its reduc-
tion might confer some mild clinical benefits.
Nevertheless, the conclusion that can be drawn
from these clinical trials is that extracellular
amyloid pathology does not necessarily me-
diate the disease’s other pathologies on the
road to dementia, in contrast to what is stip-
ulated by the linear amyloid cascade model.
With the prominence of the amyloid cascade
hypothesis, there is now a backlash, not only
against the hypothesis in general but also
against APP misprocessing as an underlying
cause of AD. Whereas this sociological dy-
namic is understandable, such a response
could be considered an overswing. Unless it
is assumed that AD’s early-onset and late-­
onset forms are mechanistically distinct,
despite being almost complete phenocopies
of each other, it would seem imprudent to throw
out APP misprocessing “with the bathwater.”
APP and its proteolytic fragments should re-
main part of any unified AD hypothesis.
So, here’s the paradox: How can APP and
its fragments be incorporated into a mecha-
nistic hypothesis of AD, despite the growing
number of clinical studies suggesting that
key assumptions of the amyloid cascade
hypothesis seem wrong.
RECONCILING THE AD PARADOX
The AD paradox can be reconciled when
viewed through the lens of 21st-century
1 of 3
 SCIENCE TRANSLATIONAL MEDICINE | FOCUS

genomics that has mapped the genomic
architecture of the sporadic late-onset form
of AD, and through other recent comple-
mentary studies that have clarified AD’s cellu-
lar and anatomical biology. Recent genetic
studies have reliably linked approximately
two dozen genes to late-onset AD. Never-
theless, this large collection of genes linked
to late-onset AD cohere around three bio-
logical pathways—endosomal trafficking,
the innate immune response, and cholester-
ol metabolism. Each pathway contains mul-
tiple genes, but they are best typified by the
genes within each pathway that have the
dependent endosomal recycling results in pathology, which can be recapitulated or rescued
slowed trafficking of proteins out of endo- by manipulating retromer-­dependent endo-
somes. This leads to enlarged and dysfunc- somal recycling [e.g., (4, 6)]. Notably, retromer
tional early endosomes (2), a “traffic jam” deficiency, of the kind observed in the post-
phenotype that occurs independently of mortem human AD brain, can lead to evi-
amyloid pathology. Endosomal enlarge- dence of progressive neurodegeneration: first
ment can also be caused by dysfunction of synaptic dysfunction (6) and then neuronal
the APP processing pathway in AD, and the cell loss [i.e., (7)]. Endosomal recycling also
cell biology of AD explains how. APP is delivers mannose-6-phosphate receptors back
primarily cleaved when it moves into early to the trans-Golgi network, where the re-
endosomal membranes, and therefore, APP’s ceptors are needed for trafficking proteases
fragments accumulate inside endosomes where from the trans-Golgi network to the lyso-
they are toxic, causing early endosome en- somes via endosomes. It is now known that
largement (3). Importantly, the traffic jam tau protein gains access to the endosomal

Downloaded from http://stm.sciencemag.org/ at CARLETON UNIVERSITY on December 2, 2020

strongest linkage to the disease. These genes phenotype is caused by APP’s C-terminal compartment, from where it is trafficked to
encode the retromer-trafficking receptor fragments, not A peptides, and thus is also the lysosomes for degradation. Recent ob-
SORL1, the microglial phagocytic receptor amyloid-independent. Hence, the two bio- servations show that manipulating retromer-­
TREM2, and the cholesterol-binding apoli- logical pathways that genetics suggests are dependent endosomal recycling mediates tau
poprotein APOE4. Among these three genes, causal—endosomal trafficking and APP pathology [e.g., (4, 5)] and can regulate its
recent studies suggest that only the gene en- processing—seem to converge more specifi- secretion (8). Consistent with their role as risk
coding SORL1 is likely to be causal for AD, cally on endosomal recycling leading to AD’s factor pathways, manipulating AD’s im-
when, in rare cases, the mutated gene causes hallmark cytopathology independent of mune response or cholesterol metabolism
loss-of-function protein truncations. Both extracellular amyloid
TREM2, whose own loss-of-function trun- deposition.
cating mutations cause other diseases but As to be expected of
not AD, and APOE4 only increase disease a causal pathway, mani­
risk and do not by themselves cause AD. pulating retromer-­ Early-onset Late-onset
These genetic observations suggest that dependent endosomal AD genes AD genes
the endosomal trafficking pathway, like the recycling in cells and
APP processing pathway, might be causal animal models can
in AD. The distinction between rare causal mediate all four of Endosomal recycling
genes (originally APP, PSEN1, PSEN2, and AD’s hallmark patho­
now likely SORL1) and common genes that logies. Endosomal re­
confer high genetic risk for AD (TREM2 cycling regulates the
and APOE4) can be debated. However, fur- outflow of APP back
ther support for the causal role of the endo- to the neuronal cell
somal trafficking pathway comes from the surface. By manipu-
cytopathology of AD and the pathophysio- lating recycling func-
logical consequences observed when proteins tion up or down, the
in each of the pathways are manipulated in endosomal recycling
cellular and animal models. pathway has been
Just as amyloid plaques and tau tangles shown t o m e d i a t e
represent AD’s hallmark histopathology, it amyloid pathology Amyloid Tau Synaptic Microglia
is now generally acknowledged that enlarged [e.g., (4, 5)], first in-
early endosomes represent its hallmark side the neuron and pathology pathology pathology pathology
CREDIT: ALICE KITTERMAN/SCIENCE TRANSLATIONAL MEDICINE

cytopathology. The endosomal trafficking
pathway begins when protein or lipid cargo
is delivered to early endosomes from the cell
surface or from the trans-Golgi network
(Fig. 1). From the early endosomes, the car-
go is either recycled back to its original
source with the aid of trafficking protein
complexes called retromers or trafficked
to late endosomes and lysosomes for deg-
radation. The enlargement of early endo-
somes in AD can be caused directly by
dysfunction of the endosomal recycling
pathway. As revealed by studies of the ret-
romer receptor SORL1, reducing retromer-­
then, through secre- Fig. 1. Proposed hub-and-spoke model for AD pathogenesis. Causal genes
tion by endosomes, in early-onset AD such as APP, PSEN1, and PSEN2 affect the intracellular processing
into the extracellular of APP and the generation of cleaved peptides including C-terminal fragments
space. Recycling also and A peptides. Putative causal genes in late-onset AD such as SORL1 affect
regulates endosomal intracellular trafficking including recycling of proteins from endosomes back to
outflow of glutamate the cell surface. Both classes of genes converge on a common pathway because
receptors at neuronal mutations in these genes result in endosomal recycling disruption, a pathogenic
synapses [e.g., (6)], a event that occurs independent of amyloid pathology. Endosomal recycling
disruption causes enlarged endosomes, a traffic jam phenotype that is a hallmark
transport route critical
cytopathology of AD. Endosomal recycling disruption can, through parallel
for neuronal plasticity pathways, trigger all four of AD’s histopathological hallmarks: extracellular and
and synaptic health. A intracellular amyloid pathology, tau neurofibrillary tangle pathology, synaptic
loss of cell-­surface gluta­ pathology, and microglia pathology. In this hub-and-spoke model, endosomal
mate receptors is charac- recycling disruption is the hub and the four downstream AD core pathologies are
teristic of AD’s synaptic the spokes.

Small and Petsko, Sci. Transl. Med. 12, eabb1717 (2020) 2 December 2020 2 of 3
SCIENCE TRANSLATIONAL MEDICINE | FOCUS
pathways can only modulate but not trig-
ger the three AD pathologies that originate
in neurons.
How about the fourth pathology that
occurs in microglia? For this pathology, the
immune response and cholesterol metabo-
lism pathways can clearly act as causal drivers,
both interacting at microglia; interestingly,
so too can endosomal recycling. Many of the
proteins of the endosomal recycling path-
way are expressed in microglia, and many
of the proteins in the immune response
pathway, such as TREM2, are phagocytic
receptors that are recycled from the endo-
somes to the microglial cell surface. It is un-
surprising, therefore, that manipulating the
retromer-dependent endosomal recycling
pathway recapitulates core features of AD’s
microglial pathology by reducing cell-­surface
phagocytic receptors in microglia [e.g., (9)].
Collectively, these studies show that
endosomal recycling can directly mediate all
four of AD’s pathologies, supporting the con-
clusion that this pathway plays a causal role
in disease. Importantly, studies have shown
that endosomal recycling can regulate tau and
synaptic pathology independent of amyloid
pathology (5, 6); this appears true also for
microglia pathology. Rather than triggering
a linear cascading series of events, endosomal
recycling defects act more like a central hub
from which each pathology fans out indepen-
dent of the others (Fig. 1). This conclusion
can in part explain why, even when extra-
cellular amyloid is cleared in patients with late-­
onset AD in relatively early stages of disease,
the other pathologies might progress unabated.
The anatomical biology of AD provides
additional support for the centrality of en-
dosomal recycling in the pathogenesis of
this disease. The entorhinal cortex, a region of
the brain’s hippocampal formation, is differ-
entially vulnerable to AD pathogenesis. Gene
expression studies designed to explain this
vulnerability have identified selective defects in
retromer-dependent endosomal recycling, in-
cluding in the retromer receptor SORL1.
With these new findings, it is now possi-
ble to propose a mechanistic model that
might resolve the AD paradox. The model
has two assumptions. First, instead of extra-
cellular amyloid pathology, intracellular
endosomal recycling disruption is proposed
to act as a primary pathogenic event that
can secondarily lead to all four of AD’s core
pathological features. Second, and perhaps
most importantly, it proposes that AD’s
four core pathologies can occur in parallel
from one central driver, giving rise to a hub-
Small and Petsko, Sci. Transl. Med. 12, eabb1717 (2020)
and-spoke model instead of a cascading
series of events (Fig. 1).
APP is incorporated into the model, be-
cause, as shown most clearly in early-onset
AD, intracellular fragments produced by
the APP processing pathway, most likely the
C-terminal fragments, can act as a primary
trigger of endosomal recycling defects (3).
In addition, we acknowledge that extracel-
lular amyloid pathology, like AD’s other
pathologies, can be detrimental and that, like
the spokes of a wheel, the downstream patho­
logies might interconnect. Nevertheless,
in the new model, amyloid pathology rep-
resents the disease’s smoke, but not its fire.
ENDOSOMAL RECYCLING: A HUB
AND SPOKE MODEL
Whereas cell and animal model systems
are useful in clarifying the mechanisms of
disease-associated pathways, the only way
to test or refute a pathogenic hypothesis of a
human disease is with patients. Just as clini-
cal trials that combined interventions for
AD with biomarkers were useful as experi-
mental studies in challenging the predictions
of extracellular amyloid pathology, so too
must clinical trials test the predictions of
disrupted endosomal recycling.
Recent preclinical studies have provided
proof-of-principle that endosomal recy-
cling disruption can be therapeutically
targeted through retromer-enhancing agents.
By increasing retromer production, these
interventions have been observed to nor-
malize endosomal recycling defects in neu-
rons including an increase in Sorl1 [e.g.,
(5)], resulting in unjamming of endosomes.
In neurons and animal models, retromer-­
enhancing agents have already been shown
to ameliorate AD’s core pathologies. They
reduce amyloid pathology and tau patholo-
gy [e.g., (4, 5)] and can ameliorate synaptic
dysfunction in the mouse brain [e.g., (6)].
How about markers of endosomal recy-
cling? A recent study suggests that an eleva-
tion of CSF tau, together with other proteins
that are known to undergo endosomal se-
cretion, occurs when retromer-dependent
endosomal recycling is disrupted (8). One
mechanistic implication of these studies is
that just like APP fragments, tau is also un-
conventionally secreted via endosomes.
Grounded in AD’s integrated biology—
that is, its genetics/molecular biology, cell
biology, and anatomical biology—the pro-
posed endosomal recycling model recon-
ciles the APP paradox and forms a plausible
2 December 2020
updated hypothesis for explaining AD patho-
genesis. Nevertheless, only when interventions
for ameliorating defective endosomal recycling
together with endosomal recycling markers are
tested in rigorous clinical trials will the ultimate
veracity of the model be confirmed or refuted.
REFERENCES AND NOTES
1. J. A. R. Nicoll, G. R. Buckland, C. H. Harrison, A. Page,
S. Harris, S. Love, J. W. Neal, C. Holmes, D. Boche,
Persistent neuropathological effects 14 years following
amyloid- immunization in Alzheimer’s disease. Brain
142, 2113–2126 (2019).
2. A. Knupp, S. Mishra, R. Martinez, J. E. Braggin, M. Szabo,
C. Kinoshita, D. W. Hailey, S. A. Small, S. Jayadev,
Downloaded from http://stm.sciencemag.org/ at CARLETON UNIVERSITY on December 2, 2020
J. E. Young, Depletion of the AD risk gene SORL1
selectively impairs neuronal endosomal traffic
independent of amyloidogenic APP processing.
Cell Rep. 31, 107719 (2020).
3. D. Kwart, A. Gregg, C. Scheckel, E. A. Murphy, D. Paquet,
M. Duffield, J. Fak, O. Olsen, R. B. Darnell, M. Tessier-Lavigne,
A large panel of isogenic APP and PSEN1 mutant
human iPSC neurons reveals shared endosomal
abnormalities mediated by APP -CTFs, not A. Neuron
104, 256–270.e5 (2019).
4. J.-G. Li, J. Chiu, D. Pratico, Full recovery
of the Alzheimer’s disease phenotype by gain
of function of vacuolar protein sorting 35.
Mol. Psychiatry 25, 2630–2640 (2019).
5. J. E. Young, L. K. Fong, H. Frankowski, G. A. Petsko,
S. A. Small, L. S. B. Goldstein, Stabilizing the retromer
complex in a human stem cell model of Alzheimer’s
disease reduces TAU phosphorylation independently
of amyloid precursor protein. Stem Cell Reports 10,
1046–1058 (2018).
6. P. Temkin, W. Morishita, D. Goswami, K. Arendt, L. Chen,
R. Malenka, The retromer supports AMPA receptor
trafficking during LTP. Neuron 94, 74–82.e5 (2017).
7. F.-L. Tang, L. Zhao, Y. Zhao, D. Sun, X.-J. Zhu, L. Mei,
W.-C. Xiong, Coupling of terminal differentiation deficit
with neurodegenerative pathology in Vps35-deficient
pyramidal neurons. Cell Death Differ. 27, 2099–2116 (2020).
8. S. Simoes, J. L. Neufeld, G. Triana-Baltzer, S. Moughadam,
E. I. Chen, M. Kothiya, Y. H. Qureshi, V. Patel, L. S. Honig,
H. Kolb, S. A. Small, Tau and other proteins found
in Alzheimer’s disease spinal fluid are linked
to retromer-mediated endosomal traffic in mice and
humans. Sci. Transl. Med. 12, eaba6334 (2020).
9. J. R. Appel, S. Ye, F. Tang, D. Sun, H. Zhang, L. Mei,
W.-C. Xiong, Increased microglial activity, impaired
adult hippocampal neurogenesis, and depressive-like
behavior in microglial VPS35-depleted mice. J. Neurosci.
38, 5949–5968 (2018).
Competing interests: G.A.P. is an adviser for Amicus
Therapeutics, MeiraGTx, Annovis Bio, and Proclara
Biosciences and holds equity or stock options in Denali
Therapeutics, MeiraGTx, Annovis Bio, and Proclara
Biosciences, companies that are developing therapies for
neurodegenerative diseases. S.A.S. holds equity or stock
options in Denali Therapeutics and is a co-inventor on a
patent that covers endosomal recycling markers. G.A.P. and
S.A.S. are co-inventors on patents covering gene therapies
for neurodegenerative diseases.
10.1126/scitranslmed.abb1717
Citation: S. A. Small, G. A. Petsko, Endosomal recycling
reconciles the Alzheimer’s disease paradox. Sci. Transl. Med.
12, eabb1717 (2020).
3 of 3
Endosomal recycling reconciles the Alzheimer's disease paradox
Scott A. Small and Gregory A. Petsko

Sci Transl Med 12, eabb1717.

DOI: 10.1126/scitranslmed.abb1717

Downloaded from http://stm.sciencemag.org/ at CARLETON UNIVERSITY on December 2, 2020

ARTICLE TOOLS http://stm.sciencemag.org/content/12/572/eabb1717

RELATED http://stm.sciencemag.org/content/scitransmed/12/571/eaba6334.full
CONTENT
http://stm.sciencemag.org/content/scitransmed/12/570/eaba1871.full
http://stm.sciencemag.org/content/scitransmed/12/563/eaaz2541.full

REFERENCES This article cites 9 articles, 2 of which you can access for free
http://stm.sciencemag.org/content/12/572/eabb1717#BIBL

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Use of this article is subject to the Terms of Service

Science Translational Medicine (ISSN 1946-6242) is published by the American Association for the Advancement of
Science, 1200 New York Avenue NW, Washington, DC 20005. The title Science Translational Medicine is a
registered trademark of AAAS.
Copyright © 2020 The Authors, some rights reserved; exclusive licensee American Association for the Advancement
of Science. No claim to original U.S. Government Works